Personalized Prediction of Vehicle Energy Consumption based on Participatory Sensing

TL;DR

This paper explores personalized vehicle energy consumption prediction using participatory sensing data, addressing data sparsity challenges and demonstrating improved accuracy through various modeling approaches.

Contribution

It introduces a systematic framework for personalized energy prediction and compares methods including blackbox modeling and collaborative filtering.

Findings

Significant improvement in prediction accuracy with proposed methods

Effective handling of missing data in participatory sensing datasets

Successful case study on electric vehicle distance-to-empty prediction

Abstract

The advent of abundant on-board sensors and electronic devices in vehicles populates the paradigm of participatory sensing to harness crowd-sourced data gathering for intelligent transportation applications, such as distance-to-empty prediction and eco-routing. While participatory sensing can provide diverse driving data, there lacks a systematic study of effective utilization of the data for personalized prediction. There are considerable challenges on how to interpolate the missing data from a sparse dataset, which often arises from participatory sensing. This paper presents and compares various approaches for personalized vehicle energy consumption prediction, including a blackbox framework that identifies driver/vehicle/environment-dependent factors and a collaborative filtering approach based on matrix factorization. Furthermore, a case study of distance-to-empty prediction for electric vehicles by participatory sensing data is conducted and evaluated empirically, which shows that our approaches can significantly improve the prediction accuracy.